A GNSS positioning method, device and storage medium based on IDR gross error detection

Through iterative data precision detection method (IDR), the abnormal and normal sets are initialized in GNSS positioning, and the observations are filtered and tested one by one, which solves multiple coarse detection difficulties, improves the accuracy and robustness of GNSS positioning, and reduces the Masking and submersion effects.

CN115390114BActive Publication Date: 2024-11-26TONGJI UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202211039658.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-08-29
Publication Date
2024-11-26
Estimated Expiration
2042-08-29

AI Technical Summary

Technical Problem

When processing geodetic data, the iterative data detection method is prone to masking and submersion effects when facing multiple coarse differences, resulting in increased difficulty in detection and identification, and statistical tests may declare more coarse differences than actual, and introduce Severe deviation.

Method used

Iterative data precision detection (IDR) is used to initialize abnormal and normal sets, and abnormal observations are filtered and tested one by one until all observations are determined to be rough or normal, reducing masking and submersion effects, and improving coarse detection and identification. Accuracy.

Benefits of technology

Effectively alleviate the masking and submersion effects, improve the accuracy and robustness of GNSS positioning, and ensure the accuracy of parameter estimation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115390114B_ABST
    Figure CN115390114B_ABST
Patent Text Reader

Abstract

The present invention relates to a GNSS positioning method, device and storage medium based on IDR gross error detection, wherein the method comprises: step 1) obtaining GNSS observation values; step 2) solving the least squares estimation of the residual variance-covariance matrix of the parameter to be estimated and the observation value; step 3) initializing a normal set and an abnormal set; step 4) determining the set to be tested, determining the w statistic, and screening the minimum w statistic; step 5) performing a significance test on the minimum w statistic, if the significance test passes, then merging the observation values ​​in the set to be tested into the normal set, updating the abnormal set and the normal set, if not, then turning to step 7); step 6) judging whether the iteration termination condition is met, if so, then turning to step 7), if not, then returning to step 4); step 7) eliminating the values ​​in the abnormal set, and solving the parameter to be estimated based on the normal set to obtain a positioning result. Compared with the prior art, the present invention has the advantages of improving the accuracy of gross error detection and the GNSS positioning accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to geodetic data processing technology, and in particular to a GNSS positioning method, device and storage medium based on IDR gross error detection. Background Art

[0002] Least Square (LS) estimation is currently the most commonly used GNSS positioning solution method. The premise of achieving unbiasedness is to use realistic function models and random models. In other words, if the observations only contain random errors, the least squares solution of the estimated parameters has the best unbiasedness. However, when there are gross errors in the observations, the accuracy of the least squares solution will be seriously affected. Therefore, before the least squares adjustment, it is crucial to detect and identify gross errors.

[0003] After more than half a century of geodesy development, data detection is one of the most mature methods for gross error detection and identification and has been used as a standard method for geodetic data quality control. Generally, data detection is based on the principle of statistical hypothesis testing and is used to detect and identify individual gross errors in linear models. It originated from the pioneering work of Baarda in 1967 and 1968, and was later extended by Pope to the case where the accuracy of the observations is unknown. The data detection method screens each individual observation to determine whether there is a gross error in it, which is the well-known w test. The test statistic used in the w test is usually the standardized or studentized least squares residual.

[0004] The data detection method first performs hypothesis testing on the null hypothesis and a set of alternative hypothesis models. In the detection process, test statistics such as normal distribution, τ distribution, χ^2 distribution, and F distribution are usually followed to determine whether there are gross errors in the observation system. Identification is a model selection, the purpose of which is to select the most trustworthy alternative hypothesis model that can correctly exclude unmodeled gross errors. Finally, adaptation or parameter estimation is performed under the identified model to eliminate or mitigate potential biases for unknown parameters. When applying the data detection method, an important prerequisite is to determine the number of suspected gross error points, which seems unrealistic in practical applications.

[0005] In modern geodesy, when the amount of data is large, there is a high probability that there are multiple gross errors in the adjustment system. In this case, the data snooping method developed theoretically for a single gross error is usually implemented successively to handle multiple gross errors by processing one gross error after another, which is called iterative data snooping (IDS). Without determining the number of gross errors in advance, continuous detection, identification and adaptation are performed under the assumption of a single gross error. Once an observation with a gross error is identified, the least squares adjustment is repeated by excluding the observation with the identified gross error. However, this method is not theoretically rigorous because it assumes that there is only one gross error in each iteration, but this assumption is immediately overturned in the next iteration. In practice, in the case of multiple gross errors, the assumption of multiple gross errors is hindered by masking and drowning effects. On the one hand, multiple gross errors can easily mask each other, increasing the difficulty of detection and identification. On the other hand, if the number of gross errors to be detected is large, statistical tests tend to declare more gross errors than there actually are. As the number of iterations increases, the probability of missed detection and false identification increases exponentially, which will introduce serious biases in parameter estimation. Summary of the invention

[0006] The purpose of the present invention is to provide a GNSS positioning method based on iterative data refinement (IDR) gross error detection, apply the w test in the data detection method to the data refining and verification process, determine an iterative data refinement method, and based on the iterative data refinement method, use the observations in the normal set to screen and verify the observations in the abnormal set one by one, put the observations in the abnormal set that are verified to be normal into the normal set, until all the observations in the abnormal set are verified to be gross errors or there are no observations in the abnormal set, so as to obtain a normal set with higher reliability, and perform gross error elimination and parameter estimation based on the normal set, effectively alleviate the masking and drowning effects, improve the accuracy of gross error detection and identification, and obtain a high-precision GNSS positioning result.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A GNSS positioning method based on IDR gross error detection comprises the following steps:

[0009] Step 1) Obtain GNSS observation values;

[0010] Step 2) solving the least squares estimate of the variance-covariance matrix of the estimated parameters and the observed value residuals based on the Gauss-Markov model;

[0011] Step 3) Initialize the normal set and abnormal set of observations based on the preconfigured maximum number of suspicious gross errors;

[0012] Step 4) determining a set to be tested, and determining a w statistic based on the least squares estimation of the variance-covariance matrix of the abnormal set, the estimated parameters, and the observed value residuals, and screening the minimum w statistic;

[0013] Step 5) Perform a significance test on the minimum w statistic. If the significance test passes, merge the observations in the set to be tested into the normal set, and update the abnormal set and the normal set. If the significance test fails, end the iteration and go to step 7);

[0014] Step 6) Determine whether the iteration termination condition is met, if so, go to step 7), if not, return to step 4) to iterate;

[0015] Step 7) Eliminate the values ​​in the abnormal set, perform parameter estimation based on the normal set and the Gauss-Markov model, solve the parameters to be estimated, and obtain a GNSS high-precision positioning result without gross error influence.

[0016] The step 2) comprises the following steps:

[0017] Step 2-1) Establish the GNSS linear observation equation based on the Gauss-Markov model

[0018] y=Ax+e,e~N(0,∑)

[0019] Where A is the m×n order non-random design matrix; x is the n×1 order parameter vector to be estimated; y is the m×1 order observation vector; e is the m×1 order accidental error; ∑ is the m×m order variance-covariance matrix that characterizes the accuracy of the observations:

[0020] ∑=σ 2 Q

[0021] Where Q is the m×m order cofactor matrix; σ 2 is the unit weight variance corresponding to the cofactor matrix;

[0022] Step 2-2) Determine the least squares solution of the parameters to be estimated:

[0023]

[0024] Step 2-3) Determine the observed residuals:

[0025]

[0026] Step 2-4) Determine the variance-covariance matrix of the estimated parameters and the observed residuals;

[0027] When the unit weight variance σ 2 When known,

[0028] The variance-covariance matrix of the estimated parameters is:

[0029]

[0030] The variance-covariance matrix of the observed residuals is:

[0031]

[0032] When the unit weight variance σ 2 When unknown,

[0033] The variance-covariance matrix of the estimated parameters is The variance-covariance matrix of the observed residuals is in,

[0034]

[0035]

[0036] is the unit weight variance σ 2 The least squares estimate of :

[0037]

[0038] The step 3) comprises the following steps:

[0039] Step 3-1) Assuming that all observations in the adjustment system form a complete set, each observation is uniquely divided into a normal set or an abnormal set based on the maximum number of suspicious gross errors q max Construct the original exception collection:

[0040]

[0041] The original normal set corresponding to the original abnormal set is recorded as:

[0042]

[0043] Step 3-2) Select the combination that satisfies the parameter solvability condition from the combination of the original abnormal-normal set, and the parameter solvability condition is the matrix or Reversible, where Is an exception collection The corresponding m×q max The design matrix is ​​of order and contains only 0 and 1 elements. Each column represents an exception set A gross error in

[0044] Step 3-3) Select a unique pair of normal-abnormal sets from the combinations that meet the parameter solvability condition as the initial normal-abnormal set to complete the initialization.

[0045] The initial normal-abnormal set obtained by screening is the jth abnormal set

[0046] When the unit weight error σ is known,

[0047]

[0048] When the unit weight error σ is unknown,

[0049]

[0050] The step 4) comprises the following steps:

[0051] Step 4-1) In the kth iteration, all observations are divided into three sets that do not overlap: the anomaly set Normal collection and the set to be tested {y i}, where y i ∈O (k-1) , O (k-1) is the set of anomalies retained in the k-1th iteration, the anomaly set The number of elements is q max -k, normal collection The number of elements is mq max +k-1, the set to be tested {y i The number of elements of} is 1, and the three sets satisfy the following two relationship conditions at the same time:

[0052]

[0053]

[0054] Step 4-2) Set of exceptions for the kth iteration Each observation in constructs the w statistic

[0055] When the unit weight error σ is known:

[0056]

[0057] When the unit weight error σ is unknown:

[0058]

[0059] in,

[0060]

[0061] is an m×(q max -k) design matrix, consisting of anomaly set The generated matrix Each column represents a set An abnormal observation in is a set of test items {y i} generated by the m×1 order design vector, is a collection of Generate m×(q max -k+1) order design matrix, that is, H i =[C i c i ];

[0062] Step 4-3) Screen the minimum w statistic:

[0063] When the unit weight error σ is known:

[0064]

[0065] When the unit weight error σ is unknown:

[0066]

[0067] The step 5) comprises the following steps:

[0068] Step 5-1) determining a significance test threshold based on a pre-configured significance level α;

[0069] Step 5-2) Compare the value of the minimum w statistic with the significance test threshold,

[0070] If the value of the minimum w statistic is less than or equal to the significance test threshold, the significance test passes, and the set to be tested {y i} j Treat it as a normal observation, put it into the normal set, and update the normal set and the abnormal set, that is:

[0071] I (k) =I (k-1) ∪{y j}

[0072] O (k) =O (k-1) -{y j}

[0073] If the value of the minimum w statistic is greater than the significance test threshold, the significance test fails, all observations in the abnormal set are regarded as abnormal observations, and the iteration ends.

[0074] When the unit weight error σ is known, the significance test threshold k α for:

[0075] k α =N α / 2 (0,1)

[0076] Among them, N α / 2 (0,1) represents the quantile value of the standard normal distribution at α / 2;

[0077] When the unit weight error σ is unknown, the significance test threshold k′ α for:

[0078] k′ α =t α / 2 (mnq max +k-1,0)

[0079] Among them, t α / 2 (mnq max +k-1,0) means the degrees of freedom are mnq max The quantile of the +k-1 student distribution at α / 2.

[0080] The iteration termination condition is that the number of iterations is equal to the maximum number of suspicious gross errors.

[0081] A GNSS positioning device based on IDR gross error detection includes a memory, a processor, and a program stored in the memory. When the processor executes the program, the method described above is implemented.

[0082] A storage medium stores a program, which implements the method described above when executed.

[0083] Compared with the prior art, the present invention has the following beneficial effects:

[0084] The present invention initializes the abnormal set and the normal set to ensure that all abnormal values ​​have been classified into the abnormal set at the beginning of iteration, and then finds the most likely normal value from the abnormal values ​​through an iterative method and verifies it to determine whether it can be classified into the normal set, thereby effectively alleviating the masking and drowning effects, ensuring the quantity and quality of observations in the normal set, and facilitating improving the accuracy of subsequent parameter estimation based on the normal set, thereby improving the positioning accuracy of the GNSS system, and having stronger robustness in the detection and identification of gross errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] Figure 1 is a flow chart of the method of the present invention;

[0086] Figure 2is the correct decision probability of IDS and IDR under the condition that σ is known;

[0087] Figure 3 This is the RMSE comparison result diagram of IDS, IDR and LS when σ is known;

[0088] Figure 4 is the correct decision probability of IDS and IDR under the condition that σ is unknown;

[0089] Figure 5 This is a comparison chart of the RMSE results of IDS, IDR and LS under the condition of unknown σ. DETAILED DESCRIPTION

[0090] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0091] The present invention applies the w test in the data detection method to the data refinement and verification process, and obtains a gross error detection method based on the Gauss-Markov model, which is called data refinement. Because its iterative form can also be used to detect and identify multiple gross errors, it is also called iterative data refinement (IDR). Based on the IDR gross error detection, a high-precision GNSS positioning method is proposed, such as Figure 1 As shown, the following steps are included:

[0092] Step 1) Obtain GNSS observation values;

[0093] Step 2) solving the least squares estimate of the variance-covariance matrix of the estimated parameters and the observed value residuals based on the Gauss-Markov model;

[0094] Step 2-1) Establish the GNSS linear observation equation based on the Gauss-Markov model

[0095] y=Ax+e,e~N(0,∑)

[0096] Where A is the m×n order non-random design matrix; x is the n×1 order parameter vector to be estimated; y is the m×1 order observation vector; e is the m×1 order accidental error; ∑ is the m×m order variance-covariance matrix that characterizes the accuracy of the observations:

[0097] ∑=σ 2 Q

[0098] Where Q is the m×m order cofactor matrix; σ 2is the unit weight variance corresponding to the cofactor matrix;

[0099] Step 2-2) Determine the least squares solution of the parameters to be estimated:

[0100]

[0101] Step 2-3) Determine the observed residuals:

[0102]

[0103] Step 2-4) Determine the variance-covariance matrix of the estimated parameters and the observed residuals;

[0104] When the unit weight variance σ 2 When known,

[0105] The variance-covariance matrix of the estimated parameters is:

[0106]

[0107] The variance-covariance matrix of the observed residuals is:

[0108]

[0109] When the unit weight variance σ 2 When unknown,

[0110] The variance-covariance matrix of the estimated parameters is The variance-covariance matrix of the observed residuals is in,

[0111]

[0112]

[0113] is the unit weight variance σ 2 The least squares estimate of :

[0114]

[0115] Step 3) Initialize the normal set and abnormal set of observations based on the preconfigured maximum number of suspicious gross errors;

[0116] Step 3-1) Assuming that all observations in the adjustment system form a complete set, each observation is uniquely divided into a normal set or an abnormal set based on the maximum number of suspicious gross errors q max Construct the original exception collection:

[0117]

[0118] The original normal set corresponding to the original abnormal set is recorded as:

[0119]

[0120] Step 3-2) Due to the special correspondence between observations and parameters, for example, some observations do not contain information about all parameters, resulting in the parameters not being completely solvable under some abnormal-normal set combinations. In this case, it is only necessary to select combinations that meet the parameter solvability conditions from the original abnormal-normal set combinations. The parameter solvability conditions are the matrix or Reversible, where Is an exception collection The corresponding m×q max The design matrix is ​​of order and contains only 0 and 1 elements. Each column represents an exception set ; for example, if the anomaly set contains the kth observation y k , then the matrix There is a k-th element in a column that is 1, and all other elements in the column are 0.

[0121] Step 3-3) Select the only most likely pair of normal-abnormal sets from the combinations that meet the parameter solvability condition as the initial normal-abnormal set to complete the initialization;

[0122] You can select the jth anomaly set As the initial anomaly set,

[0123] When the unit weight error σ is known,

[0124]

[0125] When the unit weight error σ is unknown,

[0126]

[0127] Step 4) determining a set to be tested, and determining a w statistic based on the least squares estimation of the variance-covariance matrix of the abnormal set, the estimated parameters, and the observed value residuals, and screening the minimum w statistic;

[0128] Assumptions As the initial exception set O (0) , its complement is the normal set I (0) In the following iterations, the detection and identification of gross errors can be regarded as a process of selecting normal observations from the abnormal set to the normal set.

[0129] Step 4-1) In the kth iteration, all observations are divided into three sets that do not overlap: the anomaly set Normal collection and the set to be tested {y i}, where y i ∈O (k-1) , O (k-1) is the set of anomalies retained in the k-1th iteration, the anomaly set The number of elements is q max -k, normal collection The number of elements is mq max +k-1, the set to be tested {y i The number of elements of} is 1, and the three sets satisfy the following two relationship conditions at the same time:

[0130]

[0131]

[0132] Step 4-2) Set of exceptions for the kth iteration Each observation in constructs the w statistic

[0133] When the unit weight error σ is known:

[0134]

[0135] When the unit weight error σ is unknown:

[0136]

[0137] in,

[0138]

[0139] is an m×(q max -k) design matrix, consisting of anomaly set The generated matrix Each column represents a set An abnormal observation in is a set of test items {y i} generated by the m×1 order design vector, is a collection of Generate m×(q max -k+1) order design matrix, that is, H i =[C i c i ];

[0140] Step 4-3) By screening the minimum w statistic, determine the (k-1) The most likely normal observation in

[0141] When the unit weight error σ is known:

[0142]

[0143] When the unit weight error σ is unknown:

[0144]

[0145] Step 5) Perform a significance test on the minimum w statistic. If the significance test passes, merge the observations in the set to be tested into the normal set, and update the abnormal set and the normal set. If the significance test fails, end the iteration and go to step 7);

[0146] Step 5-1) determining a significance test threshold based on a pre-configured significance level α;

[0147] When the unit weight error σ is known, the significance test threshold k α for:

[0148] k α =N α / 2 (0,1)

[0149] Among them, N α / 2 (0,1) represents the quantile value of the standard normal distribution at α / 2;

[0150] When the unit weight error σ is unknown, the significance test threshold k′ α for:

[0151] k′ α =t α / 2 (mnq max +k-1,0)

[0152] Among them, t α / 2 (mnq max +k-1,0) means the degrees of freedom are mnq max The quantile of the +k-1 student distribution at α / 2.

[0153] Step 5-2) Compare the value of the minimum w statistic with the significance test threshold,

[0154] If the value of the minimum w statistic is less than or equal to the significance test threshold, that is,

[0155]

[0156] or

[0157]

[0158] Then the significance test passes, and the set to be tested {y i} j Treat it as a normal observation, put it into the normal set, and update the normal set and the abnormal set, that is:

[0159] I (k) =I (k-1) ∪{y j}

[0160] O (k) =O (k-1) -{y j}

[0161] If the value of the minimum w statistic is greater than the significance test threshold, the significance test fails, all observations in the abnormal set are regarded as abnormal observations, and the iteration ends.

[0162] Step 6) determines whether the number of iterations is equal to the maximum number of suspicious gross errors. If so, there is no observation in the abnormal set at this time, the iteration ends and goes to step 7). If not, returns to step 4) for iteration;

[0163] Step 7) Eliminate the values ​​in the abnormal set, perform parameter estimation based on the normal set and the Gauss-Markov model, solve the parameters to be estimated, and obtain a GNSS high-precision positioning result without gross error influence.

[0164] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0165] This embodiment verifies the effectiveness of the present invention through a numerical experiment of GNSS single point positioning. The design matrix A and variance matrix ∑ are:

[0166]

[0167] ∑

[0168] =diag([8.123 7.790 8.004 14.571 8.722 6.690 8.867 6.955 6.89311.026])

[0169] The numerical experiments are carried out by Monte Carlo simulation (MCS). In each simulation, in addition to the normally distributed random errors, different numbers and sizes of gross errors are added to the observed values. Among them, the number of observations (outliers) with gross errors added is q * The number of outliers is 1 to 3, that is, the outliers are added to the observation combinations of )y1), (y1, y2) and )y1, y2, y3) respectively, and the range of the outlier size is 1 to 20 times the corresponding observation accuracy. For fairness, IDR and IDS use the same parameters, where the maximum number of suspicious outliers q max All are set to 3, and the significance level is 0.1%. For different application scenarios, two cases are considered: the known and unknown mean error of the observation unit weight σ. In order to compare and evaluate the anomaly detection and identification capabilities of the IDR and IDS methods, the following statistical analysis is performed on the various decision probability levels of multiple hypothesis tests and the root mean square error (RMSE) of the positioning solution.

[0170] Figure 2 The correct decision probabilities of IDR and IDS when σ is known are given. When there is only a single outlier, IDR

[0171] The correct decision probability of IDR is equivalent to that of IDS. When there are multiple outliers, the correct decision probability of IDR is higher than that of IDS, and the correct decision probability change trends of the two methods are significantly different. As the gross error increases, the correct decision probability of IDS increases first and then decreases, but always remains at a low level, which means that it is difficult for IDS to make a completely correct decision when facing multiple outliers. The correct decision probability of IDR will increase significantly with the increase of gross errors. It can be seen that when σ is known, for multiple outliers, IDR has stronger detection and recognition capabilities than IDS.

[0172] Figure 3 The positioning RMSE of IDR, IDS and LS in a GNSS case is given. It can be seen that when there is only a single outlier, the RMSE of LS will also increase exponentially with the increase of gross error, which means that LS

[0173] It is also not robust to a small number of outliers. At this time, both IDR and IDS can reduce the positioning error to a certain extent. However, when there are multiple outliers, the RMSE of IDS is

[0174] It will also increase seriously, even higher than LS. This shows that although IDS has been widely used in various practical applications, it may lose its robustness when facing multiple outliers. In contrast, the RMSE using IDR is much smaller and can always be controlled within a certain range. It can be seen that when σ is known, IDR has stronger robustness to multiple outliers than LS and IDS.

[0175] Next, we evaluate the ability of IDR and IDS to detect and identify outliers when σ is unknown. Figure 4 The correct decision probabilities of IDR and IDS when σ is unknown are given. It can be found that in the case of a single outlier, the correct decision probabilities of IDR and IDS are comparable, and both increase first and then remain stable as the gross error increases. For multiple outliers, due to the severe masking and drowning effects, the correct decision rate of IDS is extremely low, almost close to 0. At this time, the correct decision probability of IDR still maintains an upward trend as the gross error increases. This proves that IDR has stronger detection and recognition capabilities than IDS when dealing with multiple outliers.

[0176] Figure 5 The positioning RMSE of IDR, IDS and LS when σ is unknown are further compared. In the case of a single outlier, both methods effectively reduce the positioning error to a great extent compared to LS. At the same time, the accuracy of IDS is comparable to that of IDR. However, in the case of multiple outliers, due to the influence of masking and drowning effects, IDS has difficulty in correctly detecting and identifying outliers, resulting in a serious decrease in its robustness, and the RMSE is almost consistent with LS. In contrast, the RMSE of IDR is significantly smaller and always remains within a certain range. This shows that when σ is unknown, IDR has stronger robustness to multiple outliers compared to LS and IDS.

[0177] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A GNSS positioning method based on IDR gross error detection, characterized in that: The following steps are involved: Step 1) Obtain GNSS observation values; Step 2) solving the least squares estimate of the variance-covariance matrix of the estimated parameters and the observed value residuals based on the Gauss-Markov model; Step 3) Initialize the normal set and abnormal set of observations based on the preconfigured maximum number of suspicious gross errors; Step 4) determining a set to be tested, and determining a w statistic based on the least squares estimation of the variance-covariance matrix of the abnormal set, the estimated parameters, and the observed value residuals, and screening the minimum w statistic; Step 5) Perform a significance test on the minimum w statistic. If the significance test passes, merge the observations in the set to be tested into the normal set, and update the abnormal set and the normal set. If the significance test fails, end the iteration and go to step 7); Step 6) Determine whether the iteration termination condition is met, if so, go to step 7), if not, return to step 4) to iterate; Step 7) Eliminate the values ​​in the abnormal set, perform parameter estimation based on the normal set and the Gauss-Markov model, solve the parameters to be estimated, and obtain a GNSS high-precision positioning result without gross error influence; The step 2) comprises the following steps: Step 2-1) Establish the GNSS linear observation equation based on the Gauss-Markov model: y=Ax+e,e~N(0,∑) Where A is the m×n order non-random design matrix; x is the n×1 order parameter vector to be estimated; y is the m×1 order observation vector; e is the m×1 order accidental error; ∑ is the m×m order variance-covariance matrix that characterizes the accuracy of the observations: ∑=σ 2 Q Where Q is the m×m order cofactor matrix; σ 2 is the unit weight variance corresponding to the cofactor matrix; Step 2-2) Determine the least squares solution of the parameters to be estimated: Step 2-3) Determine the observed residuals: Step 2-4) Determine the variance-covariance matrix of the estimated parameters and the observed residuals; When the unit weight variance σ 2 When known, The variance-covariance matrix of the estimated parameters is: The variance-covariance matrix of the observed residuals is: When the unit weight variance σ 2 When unknown, The variance-covariance matrix of the estimated parameters is The variance-covariance matrix of the observed residuals is in, is the unit weight variance σ 2 The least squares estimate of : The step 3) comprises the following steps: Step 3-1) Assuming that all observations in the adjustment system form a complete set, each observation is uniquely divided into a normal set or an abnormal set based on the maximum number of suspicious gross errors qma x Construct the original exception collection: The original normal set corresponding to the original abnormal set is recorded as: Step 3-2) Select the combination that satisfies the parameter solvability condition from the combination of the original abnormal-normal set, and the parameter solvability condition is the matrix or Reversible, where Is an exception collection The corresponding design matrix is ​​m×qmax and contains only 0 and 1 elements. Each column represents an exception set A gross error in Step 3-3) Select a unique pair of normal-abnormal sets from the combinations that meet the parameter solvability condition as the initial normal-abnormal set to complete the initialization; Step 3-3) The initial normal-abnormal set obtained by screening is the jth abnormal set When the unit weight error σ is known, When the unit weight error σ is unknown, The step 4) comprises the following steps: Step 4-1) In the kth iteration, all observations are divided into three sets that do not overlap: the anomaly set Normal collection and the set to be tested {y i }, where y i ∈O (k-1) , O (k-1) is the set of anomalies retained in the k-1th iteration, the anomaly set The number of elements is q max -k, normal collection The number of elements is m-qma x +k-1, the set to be tested {y i The number of elements of} is 1, and the three sets satisfy the following two relationship conditions at the same time: Step 4-2) Exception set for the kth iteration Each observation in constructs the w statistic When the unit weight error σ is known: When the unit weight error σ is unknown: in, is an m×(q max -k) design matrix, consisting of anomaly set The generated matrix Each column represents a set An abnormal observation in is an m×1-order design vector generated by the test set {yi}, is a collection of Generate m×(q max -k+1) order design matrix, that is, H i =[C i c i ]; Step 4-3) Screen the minimum w statistic: When the unit weight error σ is known: When the unit weight error σ is unknown:

2. A GNSS positioning method based on IDR gross error detection according to claim 1, characterized in that: The step 5) comprises the following steps: Step 5-1) determining a significance test threshold based on a pre-configured significance level α; Step 5-2) Compare the value of the minimum w statistic with the significance test threshold, If the value of the minimum w statistic is less than or equal to the significance test threshold, the significance test passes, and the set to be tested {y i } j Treat it as a normal observation, put it into the normal set, and update the normal set and the abnormal set, that is: I (k) =I (k-1) ∪{y j } EITHER (k) =O (k-1) -{and j } If the value of the minimum w statistic is greater than the significance test threshold, the significance test fails, all observations in the abnormal set are regarded as abnormal observations, and the iteration ends.

3. A GNSS positioning method based on IDR gross error detection according to claim 2, characterized in that: When the unit weight error σ is known, the significance test threshold kα is: k α =N α / 2 (0,1) Among them, N α / 2 (0, 1) represents the quantile value of the standard normal distribution at α / 2; When the unit weight error σ is unknown, the significance test threshold k′ α for: k′ α =t α / 2 (m-n-q max +k-1,0) Among them, t α / 2 (mnq max +k-1,0) means the degrees of freedom are mnq max The quantile of the +k-1 student distribution at α / 2.

4. The GNSS positioning method based on IDR gross error detection according to claim 1, characterized in that: The iteration termination condition is that the number of iterations is equal to the maximum number of suspicious gross errors.

5. A GNSS positioning device based on IDR gross error detection, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 4 is implemented.

6. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 4 is implemented.